“Self-Fulfilling Prophecy” or Recognition Requires a Concept of Perception

Abstract

To the Editor: The NINDS rt-PA Stroke Trial did not prospectively assess quantity and quality of ischemic brain parenchyma as detected by baseline CT.1 In this trial, no CT reading panel graded CT scans on the basis of predefined definitions. Now the trialists hypothesized that there would be disagreement among them about the presence of early CT changes.2 They randomly selected 70 baseline scans from the trial, and 16 of them reread the scans in a 1-day session. Among the investigators was 1 neuroradiologist who served as the “gold standard.” He was able to predict the lesion location at 24 hours in 96% (95% CI 92% to 100%) of the scans based on the information provided by the baseline CT. In comparison, the 16 raters of the NINDS group were much less sensitive (78%) and specific (57%) in detecting any early CT change or in identifying changes involving >33% of the MCA territory. With this poor performance, it is not at all surprising that the agreement beyond chance among these raters was somewhat low and heterogeneous. Moreover, the authors did not design and power their study to detect an effect of the method of film viewing, the effect of image quality, or an effect of the raters’ different experiences and training on agreement. To our surprise, however, they conclude from their data that all these factors do not affect the raters’ agreement. Can we generalize this experience? Correct interpretation of CT is a problem not only for the experienced stroke specialists of the NINDS rt-PA Stroke Trial Group. Emergency …